euinc.io
Subscribe

Methodology

How status is determined

Every cell on this site is traceable to a primary source. This page describes the pipeline that gets us there, the criteria behind each status label, and the controls that keep the published record honest.

Editorial principles

  1. Primary sources first. A ministry press release ranks above a law-firm summary, which ranks above a tech-press article.
  2. Attribute everything. Every claim has a source link. No editorialised claim is published without a source.
  3. Translate but do not paraphrase original positions. Government statements are translated into English, with the original-language version one click away.
  4. Show the work. The /sources page lists every source monitored. This page documents the pipeline. The errata log below is public.
  5. No filler. A country with no news this month gets “no significant developments since {date}”. Not invented content.
  6. No takes the data does not support. Operator commentary, when it appears at all, is labeled, dated, and signed.

Status taxonomy

Each country carries two status dimensions, derived independently and surfaced on the map and on every country hero block.

Official position

  • Supportive. The government has publicly endorsed the proposal in Council or via ministry statement.
  • Constructive. The government is engaging substantively, raising specific amendments but supporting the direction.
  • Cautious. The government has raised concerns without endorsing.
  • Opposed. The government has signaled rejection or fundamental objection.
  • Quiet. No formal position has been detected.

Implementation readiness

  • Active. The national registry or ministry has begun technical or legislative preparation.
  • Signaling. Public statements about preparation exist, but no concrete action is visible.
  • Dormant. No preparation activity has been detected.

Source tiers

Each monitored source is classified into one of four tiers. Tier is the largest single input into the confidence weighting that determines whether an item auto-publishes or queues for operator review.

  • Tier A: primary government. Ministries with company-law portfolio, business registries, parliament document systems, tax-authority public guidance.
  • Tier B: practitioner. The top corporate law firms publishing client alerts in each country, plus national bar associations where they publish on EU developments.
  • Tier C: press. National business press and the national startup association where it exists.
  • Tier D: social. A curated list of policy and founder accounts on Twitter, LinkedIn, and Bluesky.

The data pipeline

A small poller fetches new content from each source on its native cadence (six hours for high-priority feeds, twenty-four hours for low). A dispatcher routes each raw item to the right OpenClaw skill based on the source's declared type: an RSS extractor, a government-press-release skill, a parliament-document skill, and so on. Each skill emits a structured signal.

The structured signal is then enriched. Translation to English uses a dedicated NMT model (NLLB-200) for cost and quality; a local LLM (Qwen 2.5 32B) handles classification, entity tagging, topic tagging, and position extraction. Anthropic's Claude is available as a fallback for genuinely tricky parses, kept on a budget cap.

A deduplication step clusters items by entity overlap and temporal proximity so a single ministry statement that gets covered across several outlets surfaces as one event with attached references.

Confidence thresholds

Each signal carries a relevance score (0 to 10) and a confidence (0 to 1). The thresholds vary by source tier:

  • Tier A items above relevance 6 and confidence 0.7 auto-publish.
  • Tier B items require relevance 7 and confidence 0.8 to auto-publish.
  • Tier C items require relevance 8 and confidence 0.85.
  • Tier D items always queue for operator review, regardless of score.

Thresholds were chosen conservatively at launch and are loosened as classifier calibration improves on observed data. When the operator overrides a status or rejects an item, the system remembers and the same source/pattern does not keep slipping through.

Composite scores

The comparison matrix exposes a small number of composite scores. Each is documented here with the exact formula, and primary data is always one click away.

  • Founder-relevance score (0 to 100). Weighted sum of implementation readiness (25 percent), EU-ESO signal favorability (25 percent), digital incorporation availability (15 percent), corporate tax favorability (15 percent), official-position supportiveness (10 percent), recent signal density (10 percent).
  • Tax attractiveness score (0 to 100). Weighted blend of CIT rate, effective rate on retained earnings, ESOP/option treatment, dividend WHT, presence of exit tax.
  • Implementation momentum score (0 to 100). Velocity of activity over the last 90 days versus the prior 90 days. Positive scores mean accelerating, negative mean decelerating.

Errata log

When something is corrected, it is logged here publicly. Empty at launch.

Report a correction by emailing samy@euinc.io. Include the URL of the page, the disputed claim, and a primary source if you have one.